Our paper on “Subcommunities of the vaginal microbiota” is out in Proceedings B
Published:
In this paper, we show that mixed membership models - here, topic models (LDA) - better describe vaginal microbiota composition than “community (state) types” that are based on clustering approaches. They also better predict the risk of transitioning to a sub-optimal state (loss of Lactobacillus dominance) and better describe longitudinal changes along menstrual cycles.
The identification of the topics rely on a package (alto) previously developed with Kris Sankaran and Julia Fukuyama and described here.
So, if you’re in need of a dimension reduction method for representing microbiota data, these described here might be a good option. Would love to discuss if you end up trying or finding something that better suit your neeeds.
I warmly thanks all my co-authors, the VMRC for their relentless efforts in understanding the vaginal microbiota and how to restore it to a healthy state, and the study participants who accepted a quite intensive collection of swabs.